CARS: A contention-aware scheduler for efficient resource management of HPC storage systems

Towards enabling I/O awareness in task-based programming models

“…Liang et al [12] proposed a contention-aware resource [14] presented an I/O batch scheduler with two policies: conservative and adaptive. The conservative policy avoids I/O congestion as much as possible targeting system-performance metrics.…”

“…Other efforts addressed I/O congestion as a global scheduling problem using global I/O aware schedulers [12], [13], [14]. The goal of this approach is to optimize whole system utilization by minimizing I/O interference between different applications running on the system.…”

Probabilistic scheduling of dynamic I/O requests via application clustering for burst‐buffers equipped high‐performance computing

“…Studies on I/O scheduling for HPC systems [9][10][11][12] are quite extensive, but relatively few on I/O scheduling for HPC platforms equipped with burst-buffers. Studies are often conducted from the viewpoint of data layout [13][14][15][16] and application coordination. 17,18 Because most applications running on HPC have obvious quasi-periodicity on I/O behavior, which alternates between computation and I/O transfer phases, 19 many studies take this application characteristic into account.…”

“…Studies on I/O scheduling for HPC systems 9–12 are quite extensive, but relatively few on I/O scheduling for HPC platforms equipped with burst‐buffers. Studies are often conducted from the viewpoint of data layout 13–16 and application coordination 17,18 …”

Programming model abstractions for optimizing I/O intensive applications

“…Liang et al [61] proposed a contention-aware resource scheduling strategy to improve the performance of the burst buffers by minimizing I/O congestion caused by I/O of different applications. This strategy analyzes I/O load on the burst buffers nodes and assigns incoming I/O to burst buffer nodes with least I/O load.…”